Handling timeouts when the request and response body size varies

[dae]

Hi there,

Firstly, thank you for all of your hard work on this crate - it is a pleasure to use.

My use case is posting requests to a remote API, where the request body and response body can vary widely, from a few kilobytes to tens of megabytes. This makes it difficult to determine an appropriate timeout for each request. I thought about setting the request timeout based on the size of the request body and some conservative estimate of a user's line speed, but it looks like the timeout that is set on the request will also apply as the response body is being read out (#502). If I've read the code correctly I could work around that by streaming the body and applying my own timeout, but I find myself wondering if there is an easier way.

Is there a way to specify a timeout on the socket read/write instead? In this case I'm not so concerned with the total amount of time taken, but simply if progress is being made or not, as that way I can avoid connections getting stuck, while not timing out unnecessarily for users with slow connections. Any suggestions you might have would be greatly appreciated.

[dae]

The hyper-timeout crate appears to enable this, but to add it in client code, reqwest would need to support a generic impl Connector being passed in, instead of hard-coding the use of Connector. Is there any interest in merging such an approach? For now I'll just use a custom version, but it would be great if this didn't require forking in the future.